Sloan School of Management MIT

MIT Sloan School of Management
Working Paper 4360-02
March 2002
PROBLEM INVESTIGATIONS IN HIGH-HAZARD
INDUSTRIES: CREATING AND NEGOTIATIONAL
LEARNING
John S. Carroll, Sachi Hatakenaka, Jenny W. Rudolph
© 2002 by John S. Carroll, Sachi Hatakenaka, Jenny W. Rudolph. All rights reserved. Short sections of text, not to
exceed two paragraphs, may be quoted without explicit permission provided that full credit including © notice is
given to the source."
This paper also can be downloaded without charge from the
Social Science Research Network Electronic Paper Collection:
http://ssrn.com/abstract_id=305719
Problem Investigations in High-Hazard Industries:
Creating and Negotiating Organizational Learning
John S. Carroll, MIT Sloan School of Management
Sachi Hatakenaka, MIT Sloan School of Management
Jenny W. Rudolph, Boston College Carroll School of Management
Summary
High-hazard or high-reliability organizations are ideal for the study of organizational
learning processes because of their intense mindfulness regarding problems. We
examine 27 problem investigation teams at 3 nuclear power plants whose task was to
report to management about causes and corrective actions and thereby contribute to
organizational learning and change. Questionnaires were given to team members and
manager recipients of the team reports, and team reports were coded regarding their
analyses and recommendations. Our results showed variable depth and creativity in
the reports, with better reports associated with more team training and experience,
and more diversity of work experience. Ratings of report quality, individual learning,
and plant changes by team members and managers suggested that reports were only
partially effective as boundary objects to reach shared understanding and negotiate
action plans. Team members rated their reports more favorably when they had better
access to information and found generic lessons for the plant and failed barriers that
could have prevented problems. Managers rated reports more favorably when the
teams had more investigation experience, better access to information, and stronger
corrective actions.
Problem Investigations in High-Hazard Industries:
Creating and Negotiating Organizational Learning 1
John S. Carroll, MIT Sloan School of Management
Sachi Hatakenaka, MIT Sloan School of Management
Jenny W. Rudolph, Boston College Carroll School of Management
Introduction
Organizational learning is a critical function for organizations, especially in rapidlychanging environments (Aldrich, 1999; Weick et al., 1999). Organizational learning
is enacted through a variety of specific practices (Popper and Lipshitz, 1998) that
support local innovation and meaning making, institutionalize new routines, and
disseminate knowledge (Crossan et al., 1999; Huber, 1991; Levitt and March, 1988).
New meaning is constructed as work activities bridge communities of practice (Cook
and Brown, 1999) or thought worlds (Dougherty, 1992) and people find new
solutions to shared problems (Schein, 1992).
In this paper, we examine a specific organizational learning practice: problem
investigations in the corrective action programs of nuclear power plants. Problem
investigations use interviews, physical inspection, and document reviews to examine
undesirable events or trends and draw lessons about underlying causes and ways to
prevent future problems. Investigations are a form of off-line reflective practice
1
Correspondence to the first author, 50 Memorial Dr., Cambridge, MA 02142.
E-mail: jcarroll@mit.edu
2
(Schon, 1983; Rudolph et al., 2000): sensemaking, analysis, and imagining of
alternatives takes place outside of the regular work process, often carried out by
individuals who were not immediately involved in the problem itself. Although
individuals investigate most problems, we focus on teams that are assigned the most
persistent, causally ambiguous, and organizationally complex problems. Problem
investigations are part of a corrective action program that starts with reporting of
problems and continues with investigation of facts and causes, generation of insights
and recommendations, implementation of interventions to improve performance, and
checking that these interventions were actually carried out with the expected results
(Schaaf et al., 1991).
Although plants devote considerable resources to investigations and
improvement programs, they may not be learning as efficiently and effectively as
they hope. Problem investigations in nuclear power plants (and other industries,
Reason, 1990) tend toward shallow analyses and reestablishment of control (rather
than learning, Sitkin et al., 1994). Investigations tend to focus on one or a few causes
that are proximal to the problem, typically involve technical faults or human agents,
have available solutions that can easily be enacted, and are acceptable to powerful
stakeholders (Carroll 1995, 1998; Rasmussen and Batstone, 1991). Further, team
learning during investigations does not automatically lead to organizational learning
and change: boundary spanning requires both willingness and effective practices
(Ancona and Caldwell, 1992; Carlile, in press; Crossan et al., 1999; Nutt, 1999;
Popper and Lipshitz, 1998).
We examine three interrelated research questions regarding the investigation
3
process and its relationship to organizational learning. We take as givens that the
team task is to prepare a report with analyses and recommendations to management,
that the team and managers may differ in their perceptions of the plant and its
problems, that the report is intended to help management engage a change
implementation process, and that the report feeds databases intended to capture
organizational learning. In this context, first, how does the composition of the team,
in terms of experience, training, and diversity of background, relate to team learning
and the quality of the team report? Second, how does team learning relate to
organizational learning and plant changes? Finally, how does the team report serve as
a boundary object (Carlile, in press; Star, 1989) negotiated between the team and
managers, a concrete and/or physical representation of knowledge that facilitates
shared practice? Before exploring these research questions, we introduce the special
challenges of learning in nuclear power plants and other high-hazard production
systems.
Learning From Failure in High-Hazard Production Systems
Nuclear power plants and other high-hazard or high-reliability production
systems (LaPorte and Consolini, 1991; Perrow, 1984) have distinct learning strategies
(Weick et al., 1999) arising from the need to understand complex interdependencies
among systems (Perrow, 1994), and avoid both potential catastrophes associated with
trial-and-error learning (Weick, 1987) and complacency that can arise in learning
only from successes (Sitkin, 1992). Weick et al. (1999) argue that maintaining high
reliability requires mindfulness comprised of attention to hazards and weak signals, a
broad action repertoire, a willingness to consider alternatives, and reluctance to
4
simplify interpretations.
Small failures, such as near-misses and incidents with minor consequences,
may be good learning opportunities because they are less risky and less threatening,
yet carry signal value for latent or hidden defects and potential chains of events that
can propagate into disasters (Reason, 1990; Sitkin, 1992). In the decades following
the accident at Three Mile Island, nuclear power plants have been encouraged by
regulators and industry groups to become aware of a larger number of minor incidents
and take action to avoid future trouble (Jackson, 1996; Rochlin, 1993). Plants that
would have formally reported a few hundred incidents each year are now routinely
reporting thousands of problems, and allocating resources to investigations on the
basis of seriousness and learning potential. This process is widely considered to have
helped the industry improve safety and reliability (e.g., Jackson, 1996), such that
serious accidents are extremely rare.
Problem Investigations and Team Reports
The task of a problem investigation team is to gather and analyze information about
the most serious and puzzling problems for the plant and to report their analyses and
recommendations to managers, who then take action. These teams innovate and learn
on behalf of the organization (Crossan et al., 1999; Dougherty and Hardy, 1996;
Huber, 1991), so greater team learning should lead to greater managerial learning and
more effective plant changes. In Figure 1, we propose a simplified temporal ordering
to these processes, starting from characteristics of the team and its access to
information, continuing to team learning and its embodiment in the team report, and
then to outcomes of the report, including management learning and change efforts
5
that are intended to improve plant performance.
*** Insert Figure 1 ***
The investigation process is a type of “delegated participation” (Nutt, 1999), a
frequently ineffective process in which representatives suggest solutions to managers
who may resist implementation for various reasons. Managers hold cultural
assumptions and values that typically emphasize results, short-term financial
objectives, and avoidance of doubt or ambiguity (Carroll, 1998; Schein, 1996;
Schulman, 1993). Teams may focus on their apparent task of finding causes (the
investigation process is usually called “Root Cause Analysis”) and offer unrealistic
recommendations since they are less aware of strategic and resource issues. Thus, it
is a considerable challenge to negotiate shared meaning between the investigation
teams and the managers who rely on the teams for insights and recommendations.
The team report is a boundary object (Carlile, in press; Star, 1989) negotiated
between the team and managers who request revisions and implement
recommendations. Boundary objects offer a concrete means for team members and
managers to specify and learn about their differences and dependencies through joint
practice that collectively transforms knowledge. But the team negotiates from a weak
power position (cf. minority influence, Wood et al., 1994) and therefore has to sell
(Dutton and Ashford, 1993) its ideas to managers who will implement change. The
result may be a watered–down report or recommendations that are not fully
implemented (Carroll et al., in press) and therefore a missed opportunity for
organizational learning.
Although the apparent function of the investigation is to issue a report that
6
contributes to a change implementation process and databases intended to capture
organizational learning, there are indirect ways that the investigation process may
enhance organizational learning. Even if managers resist immediate changes, they
may learn more about the plant and therefore be more receptive in the future.
Individual team members may carry back to their workgroups both personal
knowledge and social network links and thus enhance organizational learning
capabilities (Cook and Brown, 1999; Gruenfeld et al., 2000).
Propositions
Although we view our research as exploratory, in order to sharpen the analysis
we offer four propositions based on the concepts presented above and intended to
further articulate the research questions posed earlier. The propositions are linked to
the conceptual framework in Figure 1 by numbers referencing relationships.
Proposition 1: a good team report should present a clear description of events, offer
thoughtful analysis of underlying causes (Carroll, 1998; Reason, 1997), explore
creative avenues for learning (Morris and Moore, 2000; Sitkin et al., 1994),
recommend corrective actions that are logically connected to the analysis (Kepner
and Tregoe, 1981), and display a compelling narrative style to enhance impact.
Proposition 2: if there are differences between team members and managers in their
judgments of report quality, managers are more likely to want actionable
recommendations whereas team members are more likely to focus on discovering
causes (Carroll, 1998; Nutt, 1999).
Proposition 3: teams with more task experience (years of work, appropriate training),
more diverse functional experience (work history in more departments), and better
7
access to information will learn more and write deeper, more creative, and more
useful reports (Ancona and Caldwell, 1992; Bartunek, 1984; Jehn et al., 1999; Weick,
et al., 1999).
Proposition 4: better reports will lead to more learning by managers and more plant
change. The report captures at least some team learning, which is linked to both
organizational and individual learning (Crossan et al., 1999; Kim, 1993; Senge,
1990).
Contextual Sidebar
Industry and Organizational Context
Nuclear Power Plants and their Environment Nuclear power provides almost
20% of the electricity in the United States. During the 1990s when these data
were collected, nuclear power plants were owned and operated by electric
utilities, which typically owned other generating plants (coal, gas, hydro, etc.) and
transmission lines. Nuclear power was highly regulated with the US Nuclear
Regulatory Commission responsible for safety and separate state utility
commissions responsible for rate setting. With the spread of deregulation, electric
utilities began reorganizing to consolidate nuclear power plants into a small
number of nuclear enterprises separate from other generating sources and distinct
from the regulated transmission business.
Comment Nuclear power as a regulated industry developed a culture derived
partly from fossil power plants and their craft workers, and partly from the
nuclear navy that was the source of the technology. Plants were traditionally
8
structured with clear hierarchies and separate functional departments such as
operations, maintenance, design engineering, radiation protection, etc.
Method
Sample
Three nuclear power plants, owned and operated by three different utilities in two
different regions of the country, agreed to cooperate, based primarily on prior
contacts with the first author. One of the plants was generally considered by industry
experts to be among the leaders at conducting problem investigations. Preliminary
interviews were carried out at each plant to understand their problem investigation
and corrective action process and to identify problems within the previous two years
that had been investigated by teams.
35 problem investigation teams were identified from the prior two years, each
of which had at least 3 team members who participated broadly in team activities and
shared a perception of themselves as a team. Five teams were discarded due to
incomplete questionnaire data and two had uncodable reports that used only a
standardized report form without any causal narrative. Two investigations were
combined into one because the same team performed both simultaneously on related
problems.
Procedure
Once the teams were identified, questionnaires were distributed to team members and
shortened questionnaires went to managers who identified themselves as sponsors or
customers of the reports. Most team member questionnaires were distributed and
collected in small group sessions; the remainder were handed out with return
9
envelopes and received by mail. The questionnaire was strictly voluntary and
confidential (in most cases anonymous). Most team members took approximately
one hour to complete the questionnaire, although individuals who were members of
multiple teams took longer. The response rate from team members was 83%.
Line managers who were the customers or sponsors of each report were asked
to respond to a shortened version of the questionnaire. Since managers self-identified
as customers and distributed the questionnaires among themselves, it is not possible
to calculate a response rate. The team reports were also collected and coded.
Measures
The team member questionnaire had over 150 questions on the task, team process,
results of the report, team member cognitive style, demographics, and work history.
This paper focuses on a subset of questions relating to the quality of the report,
investigation outcomes of personal learning and plant changes, and background
information on team member demographics and work history. We defer analysis of
team ratings of team process for a later paper, except for ratings of accessibility of
information (see Appendix 1), which we consider to be the best available indicator of
the difficulty of conducting the investigation. For simplicity, we do not report
analyses of open-ended questions, although the responses contributed to our insights.
We averaged the responses from team members by team, except for departmental
affiliations, for which we counted the total number of distinct departmental
affiliations on each team, once for current assignments and again for past positions.
Managers who self-identified as customers or sponsors of each report were
asked to respond to a 33-item version of the questionnaire. For this paper, we
10
analyzed only three items also asked of the team: the quality of the report, their
personal learning, and plant changes. For teams with multiple manager responses, we
again averaged the responses by team.
For the investigation reports, we developed coding categories and procedures
through an iterative process of conceptual discussions among the research team,
coding of pilot problem reviews, and revisions using a priori and grounded
approaches (Miles and Huberman, 1994). We held several day-long meetings with
Dr. William Corcoran, an experienced industry specialist in problem investigation
programs, to review our coding protocol, code pilot problem reviews, and revise our
coding criteria. 30 codes were generated and then grouped into thematic categories:
causes, barriers/defenses, learning, corrective actions, and narrative features2. In
three rounds of pilot testing, interrater agreement improved from 60% to 88%. 23
reports were coded by two coders, one an organization theorist, the other a nuclear
engineer; differences of opinion were resolved by discussion. The organization
theorist coded four reports after the other coder left the project.
Results
Preliminary Analyses: Data Reduction
Given the small number of problem investigation teams and the large number of
items from questionnaires and report coding, we reduced the numbers of variables by
2
Causal analyses were judged on a four-point completeness scale from none to good;
specific causes, corrective actions, and narrative features were judged on a five-point
scale from none to very well; and barriers and learning were judged on a seven-point
scale that integrated completeness and explicitness of causes identified.
11
creating composite variables (see Appendix 1) based on factor analyses and scale
reliability analyses. The two questionnaire items measuring access to information
formed a moderately reliable scale (α=.54). Twelve of the 30 report codes could be
grouped into 3 dimensions (see Appendix 1): deep cause, out-of-the-box thinking,
and narrative (clear, compelling storytelling). Four of the remaining 18 individual
codes were retained that seemed important to organizational learning: failed
barriers/defenses that could have prevented problems, generic lessons learned,
corrective actions directed at fundamental problems, and corrective actions that
connected logically to causes identified. We also examined the correlations of 14
demographic and work history variables with all measures of report quality, plant
changes, and personal learning, and eliminated 4 that were never significant (age,
gender, education, and proportion of engineers/scientists). We eliminated team size
from the analysis because it was closely related to number of departments represented
on the team (r=.78, p<.001). The remaining variables, shown in Table 1, consist of
team member and manager ratings of report quality, individual learning, and plant
change, seven ratings of the team report, team ratings of information access, and nine
measures of work background. Finally, plant identity was used as a control variable
in some analyses.
*** Insert Table 1 ***
In Fig. 2 we present an informal path model of the sequence of team and
manager judgments about the report, plant changes, and their personal learning. We
assumed that the team report could be a cause of change at the plant, as intended by
the corrective action process. We further assumed that manager learning could be a
12
result of the report, and could have an impact on plant change. Finally, we placed the
team members’ own learning prior in causal sequence to the report, under the
assumption that the report captures some (but not all) of what the team members learn
individually. This causal sequence will guide our analyses of the report and its
impact on organizational learning.
*** Insert Fig. 2 ***
Team Member and Manager Ratings of the Team Report
Team members and managers judged whether, “Overall, the team produced an
excellent report.” On average, team members and managers thought the teams had
produced good reports (means of 1.80 and 2.12 with 1=strongly agree and 6=strongly
disagree), however, these judgments correlated only slightly with each other (see
Table 1, r=.18, n=27, n.s.).
Consistent with the lack of agreement in their reactions to the reports, team
ratings and manager ratings had different patterns of relationships with other
variables. More favorable team ratings (low scores were more favorable) were
significantly associated (see Table 1) with team member ratings of better access to
information (r=.64**), and our ratings of the report on finding failed barriers/defenses
(r=-.56**) and generic lessons learned (r=-.42*). In order to confirm and refine these
simple correlations, we ran a stepwise multiple regression model, with plant fixed
effects controlled by dummy variables, using access and the work background
variables. The results (see Fig. 2) showed that teams rated their reports better when
they had identified failed barriers/defenses (β=-.55***), mentioned generic lessons
learned (β=-.50***), had better access to information (β=.42***), and there were
13
more departments represented on the team (β=.32*).
In contrast, for managers, more favorable ratings of the report were associated
with corrective actions capable of addressing fundamental problems (r=-.42*) and
team experience with investigations (r=-.38*). Stepwise multiple regression analysis
controlling for plant revealed investigation experience (β=-.53**) and better access to
information (β=.43*) as significant predictors (see Fig. 2). Note that, with plant
controlled, investigation experience and corrective actions logically connected to
causes had significant partial correlations; in the stepwise regression analysis,
investigation experience entered first and suppressed logical corrective actions. Thus,
there is some evidence that managers considered corrective actions, either logically
connected to causes or directed at deeper issues, as an important aspect of the report.
Although neither relationship was significant in stepwise multiple regression analysis,
we have indicated their possible importance with dotted lines in Fig. 2 from logical
corrective actions and fundamental corrective actions to manager ratings of the report.
These results partially support Proposition 1, in that team members rated
reports significantly better that focused on failed barriers and lessons learned, and
managers tended to rate reports better that focused on corrective actions to address
problems. The differences between team members and managers are consistent with
Proposition 2. Contrary to Proposition 1, neither team members nor managers rated
reports better if the reports had a better narrative style or a deeper or more creative
analysis.
Our Coding of Characteristics of the Team Report
Our coding of the reports showed a disappointing level of depth and
14
completeness, insight and clarity. Most of the codes had means below the midpoint
on our rating scales (see Table 1). Causal analyses rarely went very deep, for
example, in one investigation of a worker who was injured falling through a roof, the
root cause was either “tunnel vision” or “failure to use Accident Prevention Manual,”
depending on which section of the report was consulted. The report did not discuss
the sources of these behaviors. Corrective actions were sometimes misaligned with
the supposed causes, either leaving root causes without corrective actions, or
introducing corrective actions without specifying which causes they would address.
Few reports were well-written -- they were sometimes confusing, often redundant,
and usually in passive voice. Nor was there an overall positive relationship among
characteristics of the report. As shown in Table 1, only about one-half of correlations
among our seven variables were positive. The strongest were relationships among
deep cause, out-of-the-box thinking, and corrective actions addressed to fundamental
causes.
Since report characteristics of failed barriers/defenses, generic lessons learned,
corrective actions capable of addressing fundamental causes, and corrective actions
corresponding to causes were related to team or manager ratings of report excellence,
we will start by analyzing their relationships with team inputs and process. As shown
in Table 1, team reports that better articulated failed barriers/defenses had team
members who had worked in more departments in the past (r=.46*) and more training
in teamwork (r=.42*). In a stepwise multiple regression model controlling for plant,
both remained significant (β =.48** and .46*, see Fig. 2). For generic lessons
learned, there were significant correlations with more years in the plant (r=.60**) and
15
fewer departments currently represented on the team (r=-.51**). This unexpected
reverse relationship with functional diversity seems to be related to differences
among plants, in that one plant had a required section of each report on lessons
learned and also tended to use smaller teams with fewer departmental affiliations. In
a stepwise regression analysis controlling for plant, as shown in Fig. 2, only years in
the plant was significant (β= .38*).
Corrective actions logically related to causes did not correlate significantly
with any of the predictors in Table 1, but with plant controlled, there was a significant
relationship with years of industry experience (β=.35*, see Fig. 2). For corrective
actions capable of addressing fundamental causes, there was a significant correlation
with more years in the industry (r=.42*). In stepwise regression analysis controlling
for plant (see Fig. 2), both years in the industry (β= .34*) and number of past
departments represented on the team (β=.35*) were significant.
Our research team also believed that an excellent report should have deep
causal analysis, out-of-the-box thinking, and a compelling narrative story. As shown
in Table 1, deep cause was correlated significantly with more analysis training
(r=.41*) and team members who had worked for more departments in the past
(r=.38*). Since neither was significant in multiple regression analysis controlling for
plant, we excluded deep cause from Fig. 2. Out-of-the-box thinking correlated
significantly with fewer years in the military (r=-.54**) and more departments
represented on the team (r=.55**). In multiple regression analysis controlling for
plant, both remained significant (β=-.36* and .30*), as shown in Fig. 2. Finally,
narrative had no significant correlations with predictor variables, but with plant
16
controlled, there was a significant relationship with training in teamwork (β=.51*).
Overall, the above results are partially consistent with Proposition 3, in that
variables related to training in teamwork, years of experience in the plant and the
industry, and number of departments represented on the team, predict characteristics
of the report. However, different measures of experience and departmental affiliation
were related to different characteristics of the report, so that the general pattern is
neither simple nor consistent. It is interesting that years in the military was
negatively related to out-of-the-box thinking, suggesting that such experience was
associated with some rigidity.
Organizational Change and Personal Learning
Both team members and managers were asked to agree or disagree with “As a result
of the team report, the plant made changes and real improvements that have addressed
the problems,” and “I personally learned a lot about the plant” from the investigation.
Team members and managers averaged between “slightly agree” and “moderately
agree” on both questions (see Table 1), with less enthusiasm than for ratings of the
report. Ratings of change by team members and managers correlate somewhat but
not significantly (r=.33 in Table 1) and ratings of personal learning correlate even less
strongly (r=.22). Interestingly, as shown in Table 1, team ratings of plant change
correlate strongly with team ratings of the report (r=.62**) but team ratings of
personal learning are uncorrelated with their ratings of the report (r=.06) or change
(r=.13). In contrast, managers’ ratings of organizational change and personal learning
are highly correlated with each other (r=.57**) and with their ratings of the report
(r=.59** and .53**). This provides qualified support for Proposition 4, the exception
17
being that personal learning by team members was not related to report quality or
plant changes. It also reinforces the separation of team learning from other aspects of
organizational learning and change.
Team ratings that the plant made changes were correlated significantly with
their ratings of the team report (r=.62**), access to information (r=.52**), finding
failed barriers/defenses (r=-.40*), better narrative (r=-.45*), having more managers or
supervisors on the team (r=-.47*), and more teamwork training (r=-.44*). In multiple
regression analysis controlling for plant, the team report (β=.45**), more managers
on the team (β=-.44**), and better narrative (β=-.31*) were significant predictors of
team ratings of plant change (see Fig. 2).
In contrast, managers’ ratings of change were correlated with their own
ratings of the team report (r=.59**) and personal learning (r=.57**), our coding of the
report having a better narrative (r=-.38*), and more team member training in
investigation techniques (r=-.48*). In multiple regression analysis controlling for
plant, reported in Fig. 2, both investigation training and rated report quality were
significant (βs=-.63** and .51*).
Team ratings that members learned from the investigation were not correlated
with any of our report coding or work characteristics. However, with plant controlled
in multiple regression analysis, teams with fewer years of plant experience reported
learning more (β=.50*). Managers’ ratings of personal learning were correlated with
managers’ rating of report quality (r=.53**) and investigation experience on the team
(r=-.55**). In multiple regression analysis controlling for plant, only prior
investigation experience was significant (β=-.57**).
18
In summary, the factors that predicted rated plant change and personal
learning were different for team members and managers. For team members, rated
change was predicted by a better report, a clear narrative in the report, and managers
on the team. This offers some support for Proposition 1 regarding the report
narrative. Having managers on the team may have helped the team sell the report to
managers, or it may have helped the team know about changes (many team members
commented in interviews and open-ended questionnaire responses that they did not
know about outcomes). Managers rated change as greater when they saw a better
report and a team with more training in root cause analysis, supporting parts of
Propositions 3 and 4. Personal learning was greater if the team members had less
plant experience, suggesting that the investigation served a training function.
Managers reported more learning when the team had more investigation experience,
again supporting parts of Propositions 3 and 4.
Discussion
Our study of problem investigation teams illustrates how use of institutionalized
organizational learning practices does not guarantee deep sensemaking and effective
change. Reports can be helpful and insightful or shallow and confusing, depending
on the skills, motivation, and support for the team. It is challenging to move from a
set of practices aimed at search, transfer, and piecemeal change toward practices that
support systemic knowledge generation among the team, managers, and the plant as a
whole (Carroll et al., in press).
Organizational learning practices require access to necessary resources
(information, training, experience), diversity of inputs and viewpoints, and the ability
19
to negotiate across disciplinary and hierarchical boundaries. In the action-oriented
work environment of nuclear power plants, it is difficult to find time to reflect, and
skills for reflection and for building shared mental models across disciplinary and
hierarchical boundaries are not widespread. In particular, there appeared to be a
disconnect between what teams thought they had learned and written into the report
and what managers had understood from the team reports and investigations.
Team Reports as Boundary Objects
We naturally assumed that the problem investigation report would be an important
part of the learning -- it is the official product of the group, the artifact that drives
corrective actions and feeds databases. We thought that a better report would have a
thorough causal analysis including fundamental causes in management, organization,
and culture; corrective actions logically tied to the causal analysis and capable of
addressing the fundamental causes; a strong learning orientation with imaginative
ideas and approaches; and a good narrative to persuasively communicate its story to a
wide audience of readers.
However, the three plants had disappointing levels of insight into their own
problems, in terms of our coding of their reports and some of their open-ended
comments and interview self-reports. Teams rarely looked for fundamental or deep,
systemic causes, but they sometimes did a good job of addressing more proximal
causes at an actionable level. Reports with more depth and creativity were associated
with teams that had more experience in the industry but less in the military and
represented more departments in their work history.
Team members and managers appeared to have learned different things from
20
the investigation and reporting process, reflecting the differences in their thought
worlds (Dougherty, 1992) or professional cultures (Schein, 1996) and the distinct
content of knowledge reservoirs (Argote and Ingram, 2000). From the teams’
viewpoint, a good report identified failed barriers and generic lessons, and was
associated with a team that had access to information. From the managers’
viewpoint, a good report came from a team with investigation experience and access
to information, and included effective corrective actions. These differences in
perspective may be effective for the overall goals of the plant, but they highlight the
challenges of using team reports as boundary objects (Carlile, in press; Star, 1989) to
enact a process of knowing (Cook and Brown, 1999) among team members and
managers that can surface mental models of the work environment, compare them,
and arrive at new, shared views.
We can offer some insights from our results into why incident investigation
reports are only partially effective as boundary objects. First, delegated participation
(Nutt, 1999) means that managers are not usually working with the team but are
waiting to respond to a draft report. Thus, the team gets some benefit of the boundary
object but negotiation is problematic with managers who have not fully participated
in the joint work. Second, the use of the boundary object to sharpen and then bridge
differences depends on rigorous attention to a shared approach. Some problems,
investigation approaches, and solutions are concrete and easy to share, but others are
not. The meaning of “root cause analysis” is not as clear and standardized as it
appears. As we noted, reports were rather casual in connecting causes and
recommended actions. It takes mindful attention to build shared understanding
21
around diffuse issues such as “culture” and “accountability” that have very different
meanings and implications to professional groups (Carroll, 1998; Carroll et al., in
press). Unless the team and managers work hard to clarify meaning and build shared
mental models, they may erode to opportunity to deepen their understanding and
produce organizational learning and change.
One outstanding positive example was a report from the well-regarded plant
that offered an explicit description of the negotiation process between the team and its
management customers. The team initially made over twenty recommendations,
which management evaluated and reduced to six, only four of which were
implemented. The report provided a uniquely candid discussion of managers’ costbenefit analyses. The reports from this plant seemed generally more explicit about
differences of opinion and specific factual arguments than reports from the other
plants, perhaps because the Plant Manager was more supportive of learning.
From Team to Organizational Learning
Organizational learning in the form of changes to routines and physical equipment
depends upon managers implementing actions as a result of the problem reports. In
general, both team members and managers reported that corrective actions had been
implemented and changes had been made that addressed the problems. For example,
in the team investigation of the worker who fell through a roof, a team member wrote,
“Management took renewed emphasis on safety. Procedures (pre-job briefs) were
changed and working aloft programs were implemented.” However, team members
frequently reported not knowing what had happened, or that management had been
defensive and therefore reports had been less than candid. One team member wrote,
22
“If top level managers aren’t willing to listen to the people doing the work, and
respond to their findings, it all becomes a waste!”
From their side of the negotiation, managers complained about long lists of
causes and corrective actions that undermined the impact of the report and seemed to
yield little value for the investment. Managers rated change as greater when the team
members had more investigation training and experience. In general, the results
reinforce the importance of both skills in teamwork and investigation and the external
function of teams -- their ability to negotiate resources and goals with management
and to establish strong relationships with other groups in order to get information and
later to sell the report (Ancona and Caldwell, 1992; Dougherty and Hardy, 1996;
Dutton and Ashford, 1993).
Implications
Our results illustrate the importance of a detailed examination of specific learning
practices. Team members and managers did not fully share an understanding of what
was learned in the report and negotiations over the report were not always effective.
Although plant personnel may wish to allocate responsibilities such that teams and
managers have different objectives and information, well-intentioned procedures for
investigations and implementation may produce disappointing benefits if there is a
failure to bridge hierarchical boundaries and communities of practice. Managers
appeared to seek logical and practical solutions to problems, not highly complex
analyses and recommendations to “solve world hunger” (a phrase from our
interviews). Deeper inquiry that could reveal new insights and systemic
understandings requires time to grapple with uncertainty and complexity and
23
therefore that managers permit themselves temporarily to “not know” and “not act”
(Schulman, 1993; Weick et al., 1999).
However, consistent with our propositions, access to information (and
boundary spanning activities in general) and team experience were both beneficial for
report quality and organizational learning. Having a multidiscipline team with
experience in many departments of the plant probably contributes to the breadth of
experience within the team and credibility with multiple audiences that enhances
access to information and buy-in to the recommendations. Having managers or
supervisors on the team also was related to team ratings of change but not managers’
ratings of change, suggesting that this may have helped give the team more
information about what was actually done as well as help bridge the boundary
between the team and managers.
From a practical standpoint, there are many implications for team composition
and training and the process of linking team and organizational learning.
Multidicipline teams bring better access to information and breadth of knowledge, but
need support to overcome misunderstandings and potential conflict (cf., Jehn et al.,
1999). Individuals and teams benefit from more training and practice, but
centralizing investigations in a specialized staff group reduces line management
participation and commitment to change (Nutt, 1999). Decentralization reduces the
direct benefit of a more expert report but increases the indirect benefits of team
members returning to their work groups with their new knowledge (e.g., those with
the least plant experience reported learning the most) and extended networks
(Gruenfeld et al., 2000). Management involvement in the investigation and in the
24
negotiation of report content must be taken seriously. At one chemical company we
visited, problem investigation activities have the explicit goal of educating managers,
not solving problems! This company makes managers collectively responsible to
understand problems in context, discuss improvement opportunities, commission
solution development activities, and implement changes.
Limitations and Future Research
We recognize that our results are more exploratory than confirmatory. With so few
teams and so many measured variables, the results need replication. Many of the
concepts that are important to our theory, such as learning and change, were
measured only indirectly by self-reports and coding of the team report. Coding of the
team report misses aspects of the learning in the conversations and reporting process
that surrounds the artifact of the report. Further, the self-report measures were
retrospective and therefore subject to bias and distortion. Future research should
incorporate direct observation and longitudinal designs. Despite the limitations,
however, the pattern of results is generally consistent, supported by open-ended
comments and interview data, and provocative for developing new connections for
theory, research, and practice.
25
Acknowledgements
This research was supported by National Science Foundation grants SBR96-1779 and
SBR98-11451 and a Marvin Bower Fellowship at the Harvard Business School. We
greatly appreciate the cooperation of the nuclear power plant sites and the staff of
their corrective action programs. Assistance in collecting, coding, analyzing, and
interpreting data was provided by Marcello Boldrini, Deborah Carroll, Christopher
Resto, Khushbu Srivastava, Annique Un, and Theodore Wiederhold. Dr. William
Corcoran provided invaluable insights into the investigation process, the nuclear
industry, and the report coding process. Deborah Ancona and Paul Carlile offered
comments on earlier drafts.
Author biographies
John S. Carroll is Professor in the Behavioral and Policy Sciences Area of the MIT
Sloan School of Management. He completed his PhD in 1973 at Harvard University.
His current research interests are decision making, organizational learning, and safety
management.
Sachi Hatakenaka is a doctoral candidate in Organizational Studies at the MIT Sloan
School of Management. Her doctoral dissertation examines university-corporation
research collaborations in the US, UK, and Japan.
Jenny W. Rudolph is a doctoral candidate in Organizational Studies at the Boston
College Carroll School of Management. Her doctoral dissertation focuses on the
sources and remediation of fixation errors by anesthesiology residents during medical
crises in a (simulated) operating room environment.
References
26
Aldrich H. 1999. Organizations Evolving. Sage: Thousand Oaks, CA.
Ancona DG, Caldwell DF. 1992. Bridging the boundary: External activity and
performance in organizational teams. Administrative Science Quarterly 37:
634-55.
Argote L. 1999. Organizational Learning: Creating, Retaining and Transferring
Knowledge. Kluwer: Norwell, MA.
Argote L, Ingram P. 2000. Knowledge transfer: A basis for competitive advantage in
firms. Organizational Behavior and Human Decision Processes 82: 150-69.
Argyris C, Schön D. 1996. Organizational Learning II: Theory, Method, and
Practice. Addison-Wesley: Reading, MA.
Bartunek JM. 1984. Changing interpretive schemes and organizational restructuring:
The example of a religious order. Administrative Science Quarterly 29: 355372.
Carlile PR. in press. A pragmatic view of knowledge and boundaries: Boundary
objects in new product development. Organization Science.
Carroll JS. 1995. Incident reviews in high-hazard industries: Sensemaking and
learning under ambiguity and accountability. Industrial and Environmental
Crisis Quarterly 9: 175-197.
Carroll JS. 1998. Organizational learning activities in high-hazard industries: The
logics underlying self-analysis. Journal of Management Studies 35: 699-717.
Carroll JS, Hatakenaka S. 2001. Driving organizational change in the midst of crisis.
MIT Sloan Management Review 42: 70-9.
27
Carroll JS, Rudolph JW, Hatakenaka S. in press. Learning from Organizational
Experience. In Handbook of Organizational Learning and Knowledge,
Easterby-Smith M, Lyles MA (eds). Blackwell:
Carroll JS, Rudolph JW, Hatakenaka S, Boldrini M. in press. The Difficult Handoff
from Incident Investigation to Implementation. Paper presented at the New
Technologies and Work Conference, Bad Homburg, Germany, May, 1999.
To be a book chapter in an edited book from the conference contributions.
Cook SDN, Brown JS. 1999. Bridging epistemologies: The generative dance between
organizational knowledge and organizational knowing. Organization Science
10: 381-400.
Crossan MM, Lane HW, White RE. 1999. An organizational learning framework:
From intuition to institution. Academy of Management Review 24: 522-37.
Dougherty D. 1992. Interpretive barriers to successful product innovation in large
firms. Organization Science 3: 179-202.
Dougherty D, Hardy C. 1996. Sustained product innovation in large, mature
organizations: Overcoming innovation-to-organization problems. Academy of
Management Journal 39: 1120-53.
Dutton JE, Ashford SJ. 1993. Selling issues to top management. Academy of
Management Review 18: 397-428.
Friedman VJ, Lipshitz R. 1992. Teaching people to shift cognitive gears:
Overcoming resistance on the road to Model II. Journal of Applied Behavioral
Science 28(1): 118-136.
28
Gruenfeld DH, Martorana PV, Fan ET. 2000. What do groups learn from their
worldiest members? Direct and indirect influence in dynamic teams.
Organizational Behavior and Human Decision Processes 82: 45-59.
Huber GP. 1991. Organizational learning: The contributing processes and the
literatures. Organization Science 2: 88-115.
Jackson SA. 1996. Challenges for the nuclear power industry and its regulators: The
NRC perspective. Speech presented at the Regulatory Information
Conference, Washington, D. C., April 9.
Jehn KA, Northcraft GB, Neale MA. 1999. Why differences make a difference: A
field study of diversity, conflict, and performance in workgroups.
Administrative Science Quarterly 44: 741-63.
Kepner CH, Tregoe BB. 1981. The New Rational Manager. Princeton: Princeton, NJ.
Kim DH. 1993. The link between individual and organizational learning. Sloan
Management Review 35: 37-50.
LaPorte TR, Consolini PM. 1991. Working in practice but not in theory: Theoretical
challenges of "High-Reliability Organizations". Journal of Public
Administration Research 1(1): 19-47.
Levitt B, March JG. 1988. Organizational learning. Annual Review of Sociology 14:
319-340.
Miles MB, Huberman AM. 1994. Qualitative Data Analysis (2nd edition). Sage:
Thousand Oaks, CA.
29
Morris MW, Moore PC. 2000. The lessons we (don’t) learn: Counterfactual thinking
and organizational accountability after a close call. Administrative Science
Quarterly 45: 737-65.
Nutt PC. 1999. Surprising but true: Half the decisions in organizations fail. Academy
of Management Executive 13: 75-90.
Perrow C. 1984. Normal Accidents. Basic Books: New York.
Popper M, Lipshitz R. 1998. Organizational learning mechanisms: A structural and
cultural approach to organizational learning. Journal of Applied Behavioral
Science 34: 161-79.
Rasmussen J, Batstone R. 1991. Toward Improved Safety Control and Risk
Management. World Bank: Washington, D.C.
Reason J. 1990. Human Error. Cambridge U.: New York.
Reason J. 1997. Managing the Risks of Organizational Accidents. Ashgate:
Brookfield, VT.
Rudolph JW, Taylor SS, Foldy EG. 2000. Collaborative off-line reflection: A way to
develop skill in action science and action inquiry. In Handbook of Action
Research, Reason P, Bradbury H (eds). Sage: Thousand Oaks.
Schaaf T Van der, Lucas DA, Hale AR. (eds). 1991. Near Miss Reporting as a Safety
Tool. Butterworth-Heinemann: Oxford.
Schein EH. 1992. Organizational culture and leadership, 2nd ed. Jossey-Bass: San
Francisco.
Schein EH. 1996. The three cultures of management: Implications for organizational
learning. Sloan Management Review 38: 9-20.
30
Schön D. 1983. The Reflective Practitioner. Basic Books: New York.
Schulman PR. 1993. The negotiated order of organizational reliability.
Administration and Society 25: 353-72.
Senge P. 1990. The Fifth Discipline. Doubleday: New York.
Sitkin SB. 1992. Learning through failure: The strategy of small losses. Research in
Organizational Behavior 14: 231-266.
Sitkin SB, Sutcliffe KM, Schroeder RG. 1994. Distinguishing control from learning
in total quality management: A contingency perspective. Academy of
Management Review 18(3): 537-64.
Star SL. 1989. The structure of ill-structured solutions: Boundary objects and
heterogeneous distributed problem solving. In Readings in Distributed
Artificial Intelligence, Huhns M, Gasser L (eds). Morgan Kaufman :Menlo
Park, CA.
Weick KE. 1987. Organizational culture as a source of high reliability. California
Management Review, Winter, 112-127.
Weick KE. 1995. Sensemaking in Organizations. Sage: Thousand Oaks, CA.
Weick KE, Sutcliffe KM, Obstfeld D. 1999. Organizing for high reliability: Processes
of collective mindfulness. Research in Organizational Beharvior 21: 81-123.
Wood W, Lundgren, S, Ouellette JA, Busceme S. Blackstone T. 1994. Minority
influence: A meta-analytic review of social influence processes.
Psychological Bulletin 115: 323-45.
31
Table 1
Descriptive Statistics and Correlations
Mean SD
1
2
3
4
5
6
7
8
9 10 11 12 13 14
1.teamreport
1.80 0.58
2.mgrreport
2.12 0.90 .18
3.teamchange
2.30 0.75 .62 .33
4.mgrchange
2.32 0.82 .23 .59 .33
5.teamlearn
2.66 0.83 .13 .10 .06 .02
6.mgrlearn
2.36 0.93 .00 .53 .20 .57 .22
7.barriers
3.24 1.53 -.56 -.16 -.40 -.27 -.09 -.19
8.fundcoract
1.85 0.55 .03 -.42 .06 -.31 .10 -.32 .37
9.logical
3.00 0.88 .03 -.32 .02 -.09 .29 -.03 .12 .22
10.generic
3.41 1.47 -.43 .08 -.19 .08 -.20 .04 -.13 -.35 -.30
11.narrative
13.14 2.03 -.25 -.22 -.45 -.38 .23 -.07 .41 .22 -.04 -.17
12.outbox
7.84 3.08 .06 -.15 .18 -.14 .09 -.05 .30 .64 .25 -.33 .21
13.deepcause 10.05 3.17 -.03 -.10 -.01 -.16 -.07 -.25 .38 .67 -.12 -.12 .33 .39
14.access
2.07 0.61 .64 .25 .52 .20 -.13 -.02 -.33 -.04 -.08 -.10 -.22 .13 -.01
15.yrsmilitary
1.42 1.25 .19 -.07 -.02 .08 -.10 -.02 -.32 -.30 -.26 .22 .24 -.54 -.03 .09
16.yrsplant
13.70 4.15 -.27 .06 -.18 .27 .23 .09 -.18 -.31 -.06 .60 -.01 -.37 -.24 -.24
17.yrsindustry 18.75 2.67 .05 -.29 -.15 .04 .30 -.14 .11 .42 .24 -.16 .15 .18 .26 -.16
18.manager
0.28 0.24 -.04 -.16 -.47 .05 -.28 -.04 .26 .08 .14 -.37 .15 .07 .11 -.14
19.depts
3.33 1.78 .32 .07 .26 .10 -.13 .19 .20 .31 .31 -.51 .05 .55 .06 .13
20.deptspast
4.70 2.05 -.23 .03 -.22 -.01 -.37 -.25 .46 .35 -.16 .03 .26 .26 .38 -.06
21.investigatns 2.35 0.61 -.03 -.38 -.18 -.33 -.31 -.55 .10 .05 -.09 .17 .01 -.10 .17 .19
22.analystrain
2.07 0.66 .01 -.14 .10 -.48 -.05 -.35 .09 .26 -.22 .16 .27 .17 .41 .19
23.teamtrain
2.38 0.57 -.34 .04 -.44 -.26 .01 -.29 .42 -.18 -.09 .29 .29 -.32 -.02 -.14
Note - correlations larger than .381 are p<.05, those larger than .487 are p<.01; variables in italics are reverse coded
32
15
.26
.13
-.07
-.32
.07
.20
.30
.18
16
17
18
19
20
21
22
.10
-.28 .38
-.47 -.11 .30
-.02 .10 .23 .33
-.13 .01 .11 -.35
-.11 .16 -.30 -.18
.32 -.00 .04 -.32
.08
.31
.14
.40
.25
.12
Figure 1
Conceptual Relationships Between Team Reports, Learning, and Change
Team
Composition
3
Team
Learning
Access to
Information
3
Report
Characteristics
3
33
1
2
Ratings of
the Report
4
Manager
Learning
4
Plant
Changes
Figure 2
Significant Paths by Stepwise Regression Analyses
Managers
.51*
TeamTrain
.38*
YrsPlant
.50*
Narrative
Generic
Team Learn
-.32*
#Dep’ts
.46*
Access
-.44**
-.50***
Barriers
.48**
-.55***
-.31*
Report (Team)
Change (Team)
.45**
.42***
RCAExper
.43*
-.53**
RCATrain
YrsIndust
.35*
-.57**
Logical
.51*
Report (Mgrs)
Mgrs’ Learn
.34*
#Dep’tsPast
.35*
FundCA
Note – italics indicate
reverse coded variables
*p<.05 **p<.01 ***p<.001
.30*
YrsMilitary
OutBox
-.36*
34
Change (Mgrs)
-.63**
Appendix 1
Composite Scales From Questionnaires and Report Coding
Composite
Access
(.54)
Narrative
(.43)
DeepCause
(.79)
Out-of-box
(.75)
Items
We were able to get all the information we needed to do our work.
Everyone the team approached gave as much of their time as the team needed.
The report tells a memorable story.
The narrative is easy to follow.
The causal chain of events leading to the incident is clear.
The causal chain of events leading to the incident is complete.
The report has an integrative explanation for various levels of causes, e.g., links
fundamental to surface causes.
The role of plant management’s priorities is mentioned
The role of plant and/or industry culture, tradition, paradigms in the incident is
examined.
The report identifies failures of oversight (by e.g. supervisors, executives, health
physics, internal quality control or external parties) that didn’t recognize
underlying weaknesses in the plant’s systems that may have led to the incident.
The report identifies previous formally processed events whose corrective actions
should have prevented this incident, i.e., failures to learn.
The report uses interdisciplinary and/or cross context approaches to researching
the problem (e.g., industry best practices, other departments in the plant).
The corrective actions suggest alternatives to existing requirements that might
rectify similar problems in the future.
The report describes the role of existing informal routines in causing the incident.
Note – team-level reliabilities are given in parentheses. The reliability for Access across
individuals was .58.
35